🌀🗞 The FLUX Review, Ep. 129

January 18th, 2024

“FCP-128” // Photo: MK with GPT-4 and DALL·E 3

Episode 129 — January 18th, 2024 — Available at read.fluxcollective.org/p/129

Contributors to this issue: Ben Mathes, Erika Rice Scherpelz, Justin Quimby, Dimitri Glazkov, Neel Mehta, Boris Smus, MK

Additional insights from: Ade Oshineye, Alex Komoroske, Robinson Eaton, Spencer Pitman, Julka Almquist, Scott Schaffter, Lisie Lillianfeld, Samuel Arbesman, Dart Lindsley, Jon Lebensold, Melanie Kahl

---

We’re a ragtag band of systems thinkers who have been dedicating our early mornings to finding new lenses to help you make sense of the complex world we live in. This newsletter is a collection of patterns we’ve noticed in recent weeks.

“Whenever a theory appears to you as the only possible one, take this as a sign that you have neither understood the theory nor the problem which it was intended to solve.”

― Karl Popper

📚🔨 A cycle of theory and practice

In theory, there’s no difference between theory and practice.

In practice, there is. 

If we do only theory or only practice, we pigeonhole ourselves. For example, we might think that hands-on work is all that’s important and that book learning is fluff — that it’s better to focus on what works right now and fix problems on the fly. Conversely, we might see big ideas and mental problem-solving as where real value comes from; hands-on work is shallow and wastes our talents. 

This is, without a doubt, a false dichotomy. When we limit ourselves to either theory or practice, we miss out on what the other side knows. If we stick only to books or hands-on stuff, we’re only getting half the story. The best solutions need synthesis. We need to practice our theories and theorize about our practices. 

We can use lessons from places like books to make hands-on work better. We can use our real-world experiences to ask new questions and develop better theories. When we mix learning with doing, we’re not just solving problems — we’re also finding new ways to think and new things to try. It’s about getting smarter and doing better all the time.

This balance is like using a map while exploring a new place. The map (theory) guides you, but walking around (practice) shows you things the map does not contain. You stumble onto a great hole-in-the-wall gelato place and know how to return to your hotel. Balanced use of different ways of knowing helps you explore more deeply than focusing on one.

This mix of thinking and doing can feel like going in circles, but it’s more of an upward spiral. The more you do, the more you learn. The more you learn, the better you get at doing. A chef improves a recipe by using what they know about flavors (theory) and adjusting based on their taste (practice). They don’t just do this once. They do it again and again to make the recipe better. 

Repeatedly alternate practice and theory: saying it out loud makes it seem almost too obvious. The hard part is applying this balance in practice. Find situations, skills, and mentors to guide you toward a good balance between theory and practice. 

🛣️🚩 Signposts 

Clues that point to where our changing world might lead us.

🚏🛍️ GPT-written Amazon listings are saying they “cannot fulfill this request” due to OpenAI’s use policy

It appears that Amazon sellers have been sloppily using LLMs to write product titles and descriptions; people on social media have been sharing pictures of items with titles containing phrases like “Apologies but I’m unable to assist with this request [as] it goes against OpenAI use policy and encourages unethical behavior,” “I cannot complete this task [as] it requires using trademarked brand names,” or simply “Sorry but I can’t provide the information you’re looking for.” Sometimes you can see instances where GPT failed to fill in a template, like in one product description that reads, “Our [product] can be used for a variety of tasks, such [task 1], [task 2], and [task 3] (sic)”.

🚏🚲 Hoboken, NJ has gone 7 years without a traffic death

Hoboken, a city in New Jersey directly across the river from Manhattan, hasn’t seen a single traffic death since January 2017; traffic injuries overall have gone down 40% in that same span, bucking the nationwide trend. The mayor’s “Vision Zero” plan has been credited with keeping Hoboken’s streets safer through a mix of simple interventions (wider curbs, high-visibility crosswalks, bike lanes) and some more sweeping changes (bikeshare programs, reduced speed limits).

🚏⛑️ Aid groups have had to suspend work in Yemen due to airstrikes on Houthis

The US and UK have begun launching airstrikes on Yemen’s Houthi rebels, who have been attacking merchant ships in the Red Sea and disrupting international commerce, but one side effect is that humanitarian aid groups have been having difficulty providing services to the residents of the war-torn country. Several organizations have suspended work due to “safety and security concerns,” which they warn could be a major problem, seeing as two-thirds of Yemenis rely on aid to survive.

🚏✈️ A battery-powered airplane design could go 500 miles on a charge

A Dutch startup has announced it’s working on a 90-seat battery-powered passenger airplane that can travel 500 miles on a single charge. Conventional wisdom has held that batteries would make planes impractically heavy, but the startup’s founders say you can work around this by rethinking the plane’s design: a narrow body and long wings improve the aircraft’s aerodynamics, and shifting batteries into the plane’s wings can make the body lighter. There are still a lot of technical challenges to work through, but the company estimates that the first flights could start in 2033.

📖⏳ Worth your time

Some especially insightful pieces we’ve read, watched, and listened to recently.

• Welcome to Hell, OpenAI (Colin Fraser) — Echoing the quips of “welcome to hell, Elon” that faced Musk when he took over Twitter, the author argues that ChatGPT got OpenAI into the unwinnable game of content moderation: no matter how you tweak the model’s parameters, somebody is going to find some output objectionable. The surface area is simply too big. The only way to make an LLM comparatively safe is to greatly restrict its possible outputs, such as an LLM that can only talk about dogs, or to constrain its inputs, such as an LLM that only lets you specify a location that you’d like to learn about.

• The Global Commons and the Hegemon (Prof. Paul Poast) — Argues that public goods (like open shipping lanes, in a modern example) are “key to a peaceful and prosperous world,” but due to the collective action problem, most countries won’t (or can’t) step up to help provide them. It requires a powerful ‘hegemon’ who is both able and willing to provide these goods, presumably because a prosperous world makes them stronger in the long run.

• ‘Ketman’ and Doublethink: What it Costs to Comply With Tyranny (Aeon) — Contra Arendt, who believed that the subjects produced by totalitarianism no longer distinguish between fact and fiction, Czesław Miłosz argued that they practiced what he called “Ketman,” first mastering deception, then practicing it competitively, valuing cunning over all else, and finally losing the ability to “differentiate his true self from the self he simulates.”

• The Essential Skills for Being Human (New York Times) — A wide-ranging, practical guide to master the art of conversation, which will help elevate people around you by seeing them reverentially, like creatures with infinite value and dignity.

🔍📆 Lens of the week

Introducing new ways to see the world and new tools to add to your mental arsenal.

This week’s lens: Schroedinger’s project.

There’s a project, and you’re not quite sure it’s making any progress. You haven’t seen anything out of it in a few weeks. So, you ask. Suddenly, there’s a flurry of activity: code submitted, emails sent, dashboards created. Was progress just not visible before… or did asking make it happen? 

We might call this a Schroedinger’s project, which is neither alive nor dead and whose state changes when you “open the box.” Often, these are projects that are kind of being worked on but also kind of not. When someone asks about them, the people responsible are reminded that they care and are motivated into a flurry of activity, either capturing and reporting the progress that was made or making that progress happen at that moment.

A Schroedinger’s project isn’t necessarily a bad thing. It’s a symptom. But a symptom of what?

For short-term efforts such as creating a new metrics dashboard or fixing an awkward but not critical bug, having a “Schroedinger” nature can act as an effective way for people to determine which of the dozens of small follow-up items on their plate are actually important. If no one cares enough to ask, it’s probably okay if it stays in the backlog forever. Imagine lazy evaluation, but for projects rather than algorithms.

Why not just prioritize each project up front and work on the top-priority ones? The reason is subtle: assigning a priority has a cost. It may involve multiple people negotiating about what is most important. A Schroedinger’s project has a yet-unresolved priority. In that uncertainty, asking about the status of a project is often implicitly understood as bumping up the priority.

However, when a longer-term effort is a Schroedinger’s project, something likely needs to change: maybe the project needs clearer ownership, a better accountability structure, more defined goals, or even deprioritization. It needs some way to tell people either that this is real and needs to make progress or that it is a low priority and can be dropped. Longer-term Schroedinger’s projects drain energy without ever realizing results.

An organization will tend to have more Schroedinger’s projects when its information processing costs are higher. Often, this is driven by the number of people involved. For instance, decentralized organizations with a highly collaborative, decentralized structure will require more informational processing to decide on shared priorities. This isn’t bad, per se, but it’s no free lunch. More collaborative work for less clear priorities is an inherent tension.

Next time you see a flurry of activity in response to a request for a status update — or, perhaps, next time you’re the one engaging in a flurry of activity in response to someone else’s query — take a moment to step back: the root cause almost always lies in the unknown, unresolved priorities. Taking the time to clarify them can help the project avoid getting back into a Schroedinger state. 

---

© 2024 The FLUX Collective. All rights reserved. Questions? Contact flux-collective@googlegroups.com.